What I learned building Harold alongside a full time job
I'm not a developer by trade. I've spent 12 years implementing ERP and business systems in manufacturing. But I realised the parts of that work I genuinely enjoyed were building and solving problems , so I decided to build something of my own.
Harold is an AI document automation tool for small businesses. The idea came from watching companies i work at manually key invoices and delivery notes into systems every day a problem enterprises solved with EDI decades ago that's never been accessible to smaller businesses.
When built the first version. I was full of ideas, features, names. It was a purely creative stage and I loved it. Then I showed it to someone outside my own head and it made no sense to them at all. I'd built it to satisfy my own creativity, not for the user. That was a real kick in the teeth moment.
That realisation sent me back to basics. I've spent far more time since then simplifying than building. This space isn't exciting. Nobody gets excited about automated data entry. It's crowded and full of promises. But I think there's an opportunity to offer the right thing at the right price self-serve, no sales calls, and customisable to how a business actually works.
That last point matters to me. Businesses are mini cultures. They know what data matters to them. For years, software companies have said "standardise to our system or pay for customisation." I think that's backwards — especially in manufacturing. A system should work for a company, not flatten what makes them unique.
What this journey has taught me so far: simplify then simplify again, don't fall in love with your own ideas, think about the user before anything else, and enjoy the small wins.
Harold is live and I'm looking for early users. If you've made the shift from implementer to builder, I'd love to know how you found it.
This is a stronger wedge than it first looks, mainly because you’re not coming at it as “AI document automation” in the abstract. You’ve seen the ugly manual workflow inside manufacturing businesses.
I’d be careful with the broad positioning though.
“Document automation for small businesses” puts Harold into a crowded AI/OCR bucket. The sharper angle is probably more specific:
Stop manually keying supplier invoices, delivery notes, and goods receipts into your accounting system.
That instantly tells the right buyer what pain you remove.
For early users, I’d probably avoid generic beta testers and go straight at small manufacturers, wholesalers, and trade businesses where someone is still forwarding PDFs, checking line items, and retyping supplier documents every week.
The strongest first-user test is not “will people try Harold?” It is:
Can you find 10 businesses still manually processing supplier documents and get 3 of them to send real examples?
That would tell you more than broad traffic.
Happy to put a tighter first-user plan in writing if useful. The useful part here is probably the exact buyer segment, outreach angle, and first 10 conversations, not more features.
Really appreciate this , the specific pain framing makes sense and honestly reflects how I've been thinking about it internally, I just haven't been sharp enough about saying it out loud.
The buyer segment is spot on. Manufacturers, wholesalers, trade businesses — that's exactly where I've spent my career and where I've watched this problem play out first hand. The person processing supplier documents isn't reading AI automation blogs, they're just doing it because nobody's fixed it for them yet.
I'd genuinely welcome the first-user plan. The 10 businesses, 3 real document conversations framing is the right goal and more useful to me right now than anything else. Would love to see how you'd approach the outreach angle specifically.
That’s exactly the right buyer.
You do not need broad beta users here. You need 10 specific conversations with people still manually handling supplier invoices, delivery notes, and goods receipts every week.
Drop your email and I’ll send over a tighter first-user plan: who to target, what to say, and how to get 3 real document examples without making it feel like a sales pitch.